What is rendering? Meaning in 3D and computer graphics

What does rendering mean in computer graphics?

Rendering is the process of computing a 2D image, or a series of frames, from the description of a 3D scene. That description holds the geometry (meshes and where they sit), the materials that say how each surface reflects or lets light through, the lights with their position, type, brightness and color, and a camera with its position, direction and focal length. The renderer reads all of it and works out a color for every pixel. The textbook Physically Based Rendering (Pharr, Jakob and Humphreys) puts it in one line: "Rendering is the process of producing an image from the description of a 3D scene."

The word covers both the act and the result. You start a rendering, and the image you get back is also a rendering, or simply a render. The program doing the work is called a renderer, a render engine or a rendering engine. For an animation, the scene description also says how objects and the camera move over time, and the renderer produces one image per frame.

In computer graphics, "rendering" on its own usually means 3D rendering. Drawing flat shapes and text, as design software and user interfaces do, is called 2D rendering, and a web browser is said to render a page when it displays text and images on screen.

Other meanings of render and rendering

Look up the meaning of rendering, or a rendering definition, in a dictionary, and the 3D sense sits among much older ones. The verb comes from Middle English rendren, from the Anglo-French rendre, "to give back, surrender", itself from the Latin reddere, according to Merriam-Webster. The senses you are most likely to meet:

• To give or provide: you render a service, render aid or render a report.
• To hand down a decision: a jury renders a verdict, a court renders a judgment.
• To make something become: new technology has rendered an old computer obsolete.
• To depict, perform or translate: a drawing is rendered in pastels, a singer gives a rendering of a song, and a new rendering of a book is a new translation.
• An architectural rendering: an image of a proposed building, either drawn by hand (watercolor, colored pencil, gouache) or generated with 3D software.
• In building: to render a wall is to coat it with plaster or cement, and that coat is called a rendering.
• In industry: rendering converts animal fat and waste tissue into stable materials such as lard or tallow.

In its artistic sense, rendering meant the work of an artist depicting a real or imaginary thing, and the finished artwork was called a rendering too. Today, to render most commonly means to have a computer generate the image from a precise description, often built by artists in 3D software.

What a renderer actually computes

Whatever the engine, the same three questions come back for every pixel.

Visibility: what does this pixel see?

The renderer first finds which surface is nearest to the camera at each pixel; everything behind it is hidden. A rasterizer does this by projecting triangles onto the screen and keeping the closest depth per pixel in a z-buffer. A ray tracer sends a ray from the camera through the pixel and computes the first object it meets. Visibility matters between surfaces too: to know whether a point is in shadow, the renderer has to know whether there is an uninterrupted path from that point to the light.

Shading: how does this surface respond to light?

Once the surface is known, its material decides what happens to the light that reaches it. Realistic renderers describe this with reflectance models (BRDFs), and textures make color and roughness vary across a surface. Some materials need more: marble, leaves and skin let light enter, scatter inside and come back out, an effect called subsurface scattering.

Light transport: where does the light come from?

A surface receives light straight from the lamps, and light that has bounced off other surfaces first. Simulating both is called global illumination, and it is what makes color bleed from one surface onto the next. On top of that come reflections, refraction through glass and water, absorption and scattering in smoke or fog, and caustics, the bright patches that appear when light is reflected or refracted before it reaches a surface. Camera effects such as depth of field and motion blur are often simulated as well. How much of this to compute exactly is a trade-off between speed, memory and realism.

The main techniques: rasterization, ray tracing, path tracing

Rendering algorithms followed a loose progression, the heavier methods becoming practical as computers got faster, and a single image can mix several of them.

Rasterization is what real-time graphics run on. Ray tracing answers the lighting questions with actual rays; the guide what is ray tracing explains how it works and how games use it.

Path tracing is a randomized form of ray tracing, proposed by Jim Kajiya in 1986 in the same paper as the rendering equation. It follows paths of light bounce after bounce and averages them, and some images need thousands of paths per pixel. It took twenty years before a feature film, Monster House, was rendered entirely with path tracing; it is now the dominant technique for realistic images, films included. Too few paths leave grain, so path tracers come with denoisers, often based on neural networks. GPUs accelerate both rasterization and ray tracing, and real-time rendering increasingly combines the two.

Real-time vs offline rendering

A basic distinction in rendering is about time. Real-time rendering generates images and displays them right away, fast enough to give the impression of motion while someone plays or moves the camera. Each frame has a fixed budget: at 30 frames per second, a frame has to be ready in under 1/30 of a second, about 33 milliseconds, and at 60 frames per second the budget drops to about 16.7 milliseconds. Real-time graphics optimize image quality within those time and hardware limits, so they lean on precomputation: lighting baked in advance for objects that never move, and light probes that record the lighting at chosen points in space for objects that do.

Offline rendering, also called pre-rendering, produces images or film frames for later viewing. Nobody is waiting for the next frame, so the renderer can use slower, more accurate methods, and a single frame may take hours or even days. Visual effects, animated films and architectural visualization are rendered this way. The budget is turned around: you decide the quality you need, and the render time follows from it.

Blender uses both ideas: all of its engines can render interactively in the 3D Viewport while you work on lighting and shading, and the final-quality image or animation is rendered once the scene is ready.

Blender's render engines: EEVEE, Cycles and Workbench

Blender ships with three render engines, picked in Render Properties > Render Engine. Each has its own settings for quality and speed, and engines from third-party developers can be added as add-ons.

EEVEE "is based on rasterization and is not a path tracer", says its manual, which adds that Cycles will always give more physically accurate renders. Both use the same shader nodes, which makes it easy to switch a scene from one to the other. Cycles renders on the CPU or, for GPU-accelerated rendering, on compatible graphics cards enabled in Edit > Preferences > System > Cycles Render Devices. Workbench is what draws the 3D Viewport in Solid mode, and the manual states it is not meant for the final images of a project.

What a render produces

Rendering the current frame gives a still image; rendering the frame range gives one image per frame. In Blender, Render > Render Image (F12) renders the active scene at the current frame, and Render > Render Animation (Ctrl+F12) renders every frame between Frame Start and Frame End, set in Output Properties > Frame Range. A still render appears in the Image Editor but is not saved on its own: you save it from there. Animation frames are written to disk as they finish, to the path set in Output Properties > Output, with the frame number added to the file name (image0001.png, image0002.png and so on).

The file format depends on what happens next. The Blender manual's rule of thumb:

• OpenEXR (16 or 32-bit float) when the images go on to compositing or color grading;
• PNG (8 or 16 bits, with alpha) for on-screen use, or to encode several video formats later;
• JPEG (8 bits, no alpha) when file size matters and some quality loss is acceptable.

For animations, the manual recommends an image sequence over a video file as soon as the total render time reaches an hour or more. A cancelled sequence can resume at the last rendered frame by changing the frame range, and the frames are then encoded to video in the Video Sequencer.

A render can also carry more than the final image (the Combined pass). Render passes, enabled in View Layer Properties > Passes, save intermediate information as separate images: depth, mist, normals, diffuse and glossy light, shadow, ambient occlusion and more. In the compositor they allow color grading, masking and relighting after the render, and the Multi-Layer EXR output keeps them all in one file.

What makes a render slow

The render time of an animation is plain arithmetic: the time of one frame multiplied by the number of frames. The Blender manual gives its own example. One minute of film at 24 frames per second is 1,440 frames, and at 30 seconds per frame that makes 720 minutes, or 12 hours of rendering. What sets the time of one frame:

• Resolution. Resolution X and Y live in Output Properties > Format. Cycles traces its samples for every pixel, so doubling both width and height means four times the pixels and roughly four times the work. The Percentage slider gives smaller test renders with the same proportions.
• Samples. In Cycles, Render Properties > Sampling sets the number of paths traced per pixel; more samples give a cleaner image and a longer render. With adaptive sampling, Cycles stops working on areas that are already clean enough, and a lower Noise Threshold means better quality and a longer render. Denoising removes grain without adding samples.
• Light bounces. Render Properties > Light Paths > Max Bounces limits how many times light bounces. Lower values render faster, and 0 gives direct lighting only. Glass usually needs the most bounces, diffuse surfaces the fewest.
• Transparency and volumes. Every transparent layer a ray crosses means running the shader and tracing one more ray. Volumes such as smoke simulations carry their own cost: the Biased method in the Cycles Volumes panel has a Max Steps limit that exists to protect from extremely long render times.

The fastest way to know where you stand is to render one typical frame at final settings, note the time and multiply; the render time calculator estimates the total from the time of one frame and the frame count. Since the frames of an animation can be rendered by different computers, a long animation can be split across many machines, which is the whole idea behind a render farm.

Rendering your first image in Blender

Three things are needed. A camera, since it defines which part of the scene is visible in the rendered image (Scene Properties > Scene > Camera picks the active one). A render engine. And a source of light: a lamp, an emissive surface or the world background. Then Render > Render Image, or F12, renders the current frame, and Render > View Render, or F11, shows the result again. Resolution, quality, output and saving the file are covered step by step in how to render in Blender.

Common mistakes and misconceptions

• Taking the viewport for the render. Solid mode is drawn by Workbench, and Material Preview by EEVEE with an HDRI environment, even in a Cycles scene. Only Rendered mode uses the scene's render engine, and in Cycles even that view has its own sample count and its own denoising switch, separate from the final render's.
• Expecting F12 to save the image. Only animation frames are saved automatically. Save a still from the Image Editor before closing Blender.
• Mixing up render and export. File > Export writes your 3D data in a format other graphics programs can process. A render writes images.
• Rendering an animation straight to a video file. The manual calls this direct approach highly not recommended. An image sequence can pick up again at the last rendered frame after an interruption.
• Timing an easy frame. The manual says to time a typical frame. A shot of empty sky underestimates the frames full of glass and smoke.
• Turning everything up to the maximum. More samples and more bounces always cost time, and the Cycles manual notes that more bounces also bring more noise. Raise a setting when the image shows a problem it fixes.